Re: [PATCH 3/3] arm64: dts: qcom: Add GPU support to X1P42100 SoC

[Akhil P Oommen <quic_akhilpo@quicinc.com>]

On 6/8/2025 8:52 PM, Dmitry Baryshkov wrote:
> On 08/06/2025 18:20, Rob Clark wrote:
>> On Sun, Jun 8, 2025 at 8:09 AM Dmitry Baryshkov
>> <dmitry.baryshkov@oss.qualcomm.com> wrote:
>>>
>>> On Sun, Jun 08, 2025 at 07:10:11AM -0700, Rob Clark wrote:
>>>> On Sat, Jun 7, 2025 at 1:17 PM Dmitry Baryshkov
>>>> <dmitry.baryshkov@oss.qualcomm.com> wrote:
>>>>>
>>>>> On Sat, Jun 07, 2025 at 07:45:01PM +0530, Akhil P Oommen wrote:
>>>>>> X1P42100 SoC has a new GPU called Adreno X1-45 which is a smaller
>>>>>> version of Adreno X1-85 GPU. Describe this new GPU and also add
>>>>>> the secure gpu firmware path that should used for X1P42100 CRD.
>>>>>>
>>>>>> Signed-off-by: Akhil P Oommen <akhilpo@oss.qualcomm.com>
>>>>>> ---
>>>>>>   arch/arm64/boot/dts/qcom/x1e80100.dtsi    |   7 ++
>>>>>>   arch/arm64/boot/dts/qcom/x1p42100-crd.dts |   4 +
>>>>>>   arch/arm64/boot/dts/qcom/x1p42100.dtsi    | 121 ++++++++++++++++
>>>>>> +++++++++++++-
>>>>>>   3 files changed, 131 insertions(+), 1 deletion(-)
>>>>>>
>>>>>> diff --git a/arch/arm64/boot/dts/qcom/x1e80100.dtsi b/arch/arm64/
>>>>>> boot/dts/qcom/x1e80100.dtsi
>>>>>> index
>>>>>> a8eb4c5fe99fe6dd49af200a738b6476d87279b2..558d7d387d7710770244fcc901f461384dd9b0d4 100644
>>>>>> --- a/arch/arm64/boot/dts/qcom/x1e80100.dtsi
>>>>>> +++ b/arch/arm64/boot/dts/qcom/x1e80100.dtsi
>>>>>> @@ -8245,6 +8245,13 @@ sbsa_watchdog: watchdog@1c840000 {
>>>>>>                        interrupts = <GIC_SPI 0 IRQ_TYPE_LEVEL_HIGH>;
>>>>>>                };
>>>>>>
>>>>>> +             qfprom: efuse@221c8000 {
>>>>>> +                     compatible = "qcom,x1e80100-qfprom",
>>>>>> "qcom,qfprom";
>>>>>> +                     reg = <0 0x221c8000 0 0x1000>;
>>>>>> +                     #address-cells = <1>;
>>>>>> +                     #size-cells = <1>;
>>>>>> +             };
>>>>>> +
>>>>>>                pmu@24091000 {
>>>>>>                        compatible = "qcom,x1e80100-llcc-bwmon",
>>>>>> "qcom,sc7280-llcc-bwmon";
>>>>>>                        reg = <0 0x24091000 0 0x1000>;
>>>>>> diff --git a/arch/arm64/boot/dts/qcom/x1p42100-crd.dts b/arch/
>>>>>> arm64/boot/dts/qcom/x1p42100-crd.dts
>>>>>> index
>>>>>> cf07860a63e97c388909fb5721ae7b9729b6c586..cf999c2cf8d4e0af83078253fd39ece3a0c26a49 100644
>>>>>> --- a/arch/arm64/boot/dts/qcom/x1p42100-crd.dts
>>>>>> +++ b/arch/arm64/boot/dts/qcom/x1p42100-crd.dts
>>>>>> @@ -15,3 +15,7 @@ / {
>>>>>>        model = "Qualcomm Technologies, Inc. X1P42100 CRD";
>>>>>>        compatible = "qcom,x1p42100-crd", "qcom,x1p42100";
>>>>>>   };
>>>>>> +
>>>>>> +&gpu_zap_shader {
>>>>>> +     firmware-name = "qcom/x1p42100/gen71500_zap.mbn";
>>>>>> +};
>>>>>> diff --git a/arch/arm64/boot/dts/qcom/x1p42100.dtsi b/arch/arm64/
>>>>>> boot/dts/qcom/x1p42100.dtsi
>>>>>> index
>>>>>> 27f479010bc330eb6445269a1c46bf78ec6f1bd4..5ed461ed5cca271d43647888aa6eacac3de2ac9d 100644
>>>>>> --- a/arch/arm64/boot/dts/qcom/x1p42100.dtsi
>>>>>> +++ b/arch/arm64/boot/dts/qcom/x1p42100.dtsi
>>>>>> @@ -17,15 +17,134 @@
>>>>>>   /delete-node/ &cpu_pd9;
>>>>>>   /delete-node/ &cpu_pd10;
>>>>>>   /delete-node/ &cpu_pd11;
>>>>>> +/delete-node/ &gpu_opp_table;
>>>>>>   /delete-node/ &pcie3_phy;
>>>>>>
>>>>>>   &gcc {
>>>>>>        compatible = "qcom,x1p42100-gcc", "qcom,x1e80100-gcc";
>>>>>>   };
>>>>>>
>>>>>> -/* The GPU is physically different and will be brought up later */
>>>>>> +&gmu {
>>>>>> +     /delete-property/ compatible;
>>>>>> +     compatible = "qcom,adreno-gmu-x145.0", "qcom,adreno-gmu";
>>>>>> +};
>>>>>> +
>>>>>> +&qfprom {
>>>>>> +     gpu_speed_bin: gpu_speed_bin@119 {
>>>>>> +             reg = <0x119 0x2>;
>>>>>> +             bits = <7 9>;
>>>>>> +     };
>>>>>> +};
>>>>>> +
>>>>>>   &gpu {
>>>>>>        /delete-property/ compatible;
>>>>>
>>>>> I think, you can drop this line.

I wasn't sure about this and I thought it was harmless to delete it.
Anyway, I will remove the "delete" from both GPU and GMU nodes.

>>>>>
>>>>>> +
>>>>>> +     compatible = "qcom,adreno-43030c00", "qcom,adreno";
>>>>>> +
>>>>>> +     nvmem-cells = <&gpu_speed_bin>;
>>>>>> +     nvmem-cell-names = "speed_bin";
>>>>>> +
>>>>>> +     gpu_opp_table: opp-table {
>>>>>> +             compatible = "operating-points-v2-adreno",
>>>>>> "operating-points-v2";
>>>>>> +
>>>>>> +             opp-1400000000 {
>>>>>> +                     opp-hz = /bits/ 64 <1400000000>;
>>>>>> +                     opp-level = <RPMH_REGULATOR_LEVEL_TURBO_L4>;
>>>>>> +                     opp-peak-kBps = <16500000>;
>>>>>> +                     qcom,opp-acd-level = <0xa8295ffd>;
>>>>>> +                     opp-supported-hw = <0x3>;
>>>>>> +             };
>>>>>> +
>>>>>> +             opp-1250000000 {
>>>>>> +                     opp-hz = /bits/ 64 <1250000000>;
>>>>>> +                     opp-level = <RPMH_REGULATOR_LEVEL_TURBO_L3>;
>>>>>> +                     opp-peak-kBps = <16500000>;
>>>>>> +                     qcom,opp-acd-level = <0x882a5ffd>;
>>>>>> +                     opp-supported-hw = <0x7>;
>>>>>> +             };
>>>>>> +
>>>>>> +             opp-1107000000 {
>>>>>> +                     opp-hz = /bits/ 64 <1107000000>;
>>>>>> +                     opp-level = <RPMH_REGULATOR_LEVEL_TURBO_L1>;
>>>>>> +                     opp-peak-kBps = <16500000>;
>>>>>> +                     qcom,opp-acd-level = <0x882a5ffd>;
>>>>>> +                     opp-supported-hw = <0xf>;
>>>>>> +             };
>>>>>> +
>>>>>> +             opp-1014000000 {
>>>>>> +                     opp-hz = /bits/ 64 <1014000000>;
>>>>>> +                     opp-level = <RPMH_REGULATOR_LEVEL_TURBO>;
>>>>>> +                     opp-peak-kBps = <14398438>;
>>>>>> +                     qcom,opp-acd-level = <0xa82a5ffd>;
>>>>>> +                     opp-supported-hw = <0xf>;
>>>>>> +             };
>>>>>> +
>>>>>> +             opp-940000000 {
>>>>>> +                     opp-hz = /bits/ 64 <940000000>;
>>>>>> +                     opp-level = <RPMH_REGULATOR_LEVEL_NOM_L1>;
>>>>>> +                     opp-peak-kBps = <14398438>;
>>>>>> +                     qcom,opp-acd-level = <0xa82a5ffd>;
>>>>>> +                     opp-supported-hw = <0xf>;
>>>>>> +             };
>>>>>> +
>>>>>> +             opp-825000000 {
>>>>>> +                     opp-hz = /bits/ 64 <825000000>;
>>>>>> +                     opp-level = <RPMH_REGULATOR_LEVEL_NOM>;
>>>>>> +                     opp-peak-kBps = <12449219>;
>>>>>> +                     qcom,opp-acd-level = <0x882b5ffd>;
>>>>>> +                     opp-supported-hw = <0xf>;
>>>>>> +             };
>>>>>> +
>>>>>> +             opp-720000000 {
>>>>>> +                     opp-hz = /bits/ 64 <720000000>;
>>>>>> +                     opp-level = <RPMH_REGULATOR_LEVEL_SVS_L2>;
>>>>>> +                     opp-peak-kBps = <10687500>;
>>>>>> +                     qcom,opp-acd-level = <0xa82c5ffd>;
>>>>>> +                     opp-supported-hw = <0xf>;
>>>>>> +             };
>>>>>> +
>>>>>> +             opp-666000000-0 {
>>>>>> +                     opp-hz = /bits/ 64 <666000000>;
>>>>>> +                     opp-level = <RPMH_REGULATOR_LEVEL_SVS_L1>;
>>>>>> +                     opp-peak-kBps = <8171875>;
>>>>>> +                     qcom,opp-acd-level = <0xa82d5ffd>;
>>>>>> +                     opp-supported-hw = <0xf>;
>>>>>> +             };
>>>>>> +
>>>>>> +             /* Only applicable for SKUs which has 666Mhz as Fmax */
>>>>>> +             opp-666000000-1 {
>>>>>> +                     opp-hz = /bits/ 64 <666000000>;
>>>>>> +                     opp-level = <RPMH_REGULATOR_LEVEL_SVS_L1>;
>>>>>> +                     opp-peak-kBps = <16500000>;
>>>>>
>>>>> This looks odd, why is it so high?
>>>>
>>>> You want max bandwidth on max opp
>>>
>>> Yes, but can it actually sustain / provide this BW?
>>>
>>
>> I'd have to trust Akhil on that one, but I have no reason to believe
>> otherwise.  Just pointing out we've done analogous things elsewhere
>> (for ex, cpu bw for sc7180-lite.dtsi)

Window's GPU KMD seems to vote Max bandwidth for this SKU, so I think
this is fine. Our GPUs from last few generations can easily saturate DDR
with the right usecase.

-Akhil

> 
> Ack. Then I'll wait for v2 with no deleting of compatible lines (a new
> line here would just replace the existing one).
> 
>